運用資料探勘於自動化色彩語意分析之研究

Abstract

電腦科技與資料科學的發展，促使色彩分析方式產生轉變，運用機器學習理論解讀色彩語意。透過跨領域技術整合，將色彩語意分析與實務應用引入更有效率、更低成本的分析方法，為本研究最具價值之處。本研究提出一個全新的色彩語意分析方法，配合網路大數據、卷積神經網路以及改良式中位切割演算法等資料探勘方式，分析詞彙的色彩意象，得出詞彙具體的對應RGB值；再依據調和配色理論，自動產出配色組合。最後透過問卷調查，評估色彩語意分析方法實際應用的可行性。 研究結果顯示，本研究提出之色彩語意分析方法，符合過去文獻與問卷調查之結果，並能找到詞彙對應之色彩趨勢，可省卻心理實驗的時間與人力成本，並且更有彈性。透過將調和配色理論數值化定義，產出之兩組配色應用設計（相似配色、補色配色），不僅能在短時間內產出大量色彩組合，且相較於人們依照直覺的配色，此方法更為客觀。將兩組配色應用設計之色彩組合與配色網站Adobe CC使用者配色比較詞彙之間的符合程度，其中一組符合程度稍差，另一組符合程度則與使用者提出的色彩組合相近，顯示此配色方法雖尚不及人為的配色操作，但仍具有極大發展與進步空間。The methods of color analysis have changed because of the development of computer technology and data science; many studies use data mining techniques to interpret the color image of semantics. The most valuable aspect of this research is introduce new technologies to enable color semantic analysis in a more efficient and cost-effective way. This study proposes a new color semantic analysis method, which is combined with big data from internet, convolutional neural network and modified median cut quantization to analyze the color image of semantic and obtain the corresponding RGB value of vocabulary. By formulating a quantization algorithm based on the rules of colour harmony, the color combination is generated automatically. Finally, the feasibility of the color semantic analysis method is evaluated through questionnaire survey. The results show that the analysis method proposed in this study can meet the results of past literature and questionnaire survey, and find the color trend corresponding to the vocabulary. By formulating quantization algorithm based on the rules of color harmony, we produce two sets of color combinations as samples. Compared the degree of conformity with human samples, one of our sample group is slightly worse, and the other sample group is similar to the color combination proposed by human, more than half of the people think that the samples and vocabularies feel in line with each other.